Vision Research 46 (2006) 3886–3897www.elsevier.com/locate/visres

The spatio-temporal tuning of the mechanisms in the control of saccadic eye movements

Brian J. White a,¤, Dirk Kerzel b, Karl R. Gegenfurtner a

a Justus-Liebig-Universität Giessen, Allgemeine Psychologie, Otto-Behaghel-Str. 10F, D-35394 Giessen, Germanyb Université de Genève, Uni Mail, FaPSE 40 bd du Pont d’Arve, CH-1205 Genève, Switzerland

Received 2 March 2006; received in revised form 16 June 2006

Abstract

We compared the spatio-temporal tuning of perception to the mechanisms that drive saccadic eye movements. Detection thresholdswere measured for Gabor-targets presented left or right of Wxation (4 or 8 deg eccentricity), at one of four spatial frequencies (1, 2, 4 or8 cpd) oscillating at one of three temporal frequencies (1, 8 or 16 Hz). We then measured saccade latency to each target presented at vari-ous multiples of detection threshold. Consistent with previous research, latency decreased as a function of contrast. However, at equaldetection performance, we found no systematic diVerence in saccadic latency and no diVerence in average oculometric performance(% correct saccade direction) across the diVerent target spatio-temporal frequencies. Furthermore, position error remained fairly constantacross all conditions. The results are consistent with the idea that the spatio-temporal signals used for perception are the same as thoseused by the mechanisms driving saccadic eye movements.© 2006 Elsevier Ltd. All rights reserved.

Keywords: Saccade latency; Spatio-temporal tuning; Temporal frequency; Spatial frequency; Contrast sensitivity

1. Introduction

The patterns of light that enter our eyes during everydayvision are Wlled with spatio-temporal variation. Because offactors related to both the optics of the eye (Williams, Brai-nard, McMahon, & Navarro, 1994) and the brain, contrastsensitivity varies as a function of spatio-temporal frequency(e.g., Kelly, 1979) and retinal eccentricity (e.g., Rovamo,Virsu, & Nasanen, 1978; Wright & Johnston, 1983). Thespatio-temporal tuning of neurons at diVerent levels ofvisual processing is well established (e.g., De Valois, Albr-echt, & Thorell, 1982; Derrington & Lennie, 1984; Merigan& Maunsell, 1993). However, much less is known about thenature of the visual signals driving the motor system. Interms of saccades, the superior colliculus (SC) and frontaleye Welds (FEF) depend on inputs from visual areas that

* Corresponding author. Fax: +49 641 9926119.E-mail address: brian.j.white@psychol.uni-giessen.de (B.J. White).

0042-6989/$ - see front matter © 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.visres.2006.06.012

convey selectivity for features like orientation and color,since visually responsive neurons in the SC and FEF arenot believed to be selective for visual features (Bruce &Goldberg, 1985; Robinson & McClurkin, 1989). The ques-tion this paper addresses is to what degree do the visuosacc-adic and perceptual systems use the same spatio-temporalsignals.

The delay between the appearance of a peripheral targetand the onset of a saccade (saccade latency) represents thesum of the visuomotor processes underlying the movement(Findlay & Gilchrist, 2003). Along these lines, saccadelatency has been modeled as a decision signal that rises tosome response threshold (LATER Model, Carpenter,1988), at a rate that is inXuenced by both goal-driven (priorexpectations) and stimulus-driven (accumulation ofsensory information) factors, the latter of which is the focushere.

Saccade latency has been of considerable interest forseveral reasons: It can reXect underlying neuronal latencies,and can be indicative of diVerent mechanisms driving the

B.J. White et al. / Vision Research 46 (2006) 3886–3897 3887

movements (e.g., “express saccades” as revealed by abimodal latency distribution, see Fischer & Ramsperger,1984). Latencies have also been used to examine the inXu-ence of distractors (Trappenberg, Dorris, Munoz, & Klein,2001; Walker, Deubel, Schneider, & Findlay, 1997; Weber& Fischer, 1994; White, Gegenfurtner, & Kerzel, 2005), andthe competition between stimulus-driven (“bottom-up”)and goal-driven (“top-down”) control (Ludwig & Gilchrist,2002). Furthermore, latencies can be used to investigate thesensitivity of the oculomotor system to various target prop-erties such as contrast (Doma & Hallett, 1988), retinaleccentricity (Kalesnykas & Hallett, 1994), and spatial fre-quency (SF) (Ludwig, Gilchrist, & McSorley, 2004). Forexample, Ludwig et al. (2004) reported that, at any givenmultiple of contrast threshold, targets of low spatial fre-quency (1–2 cpd) triggered saccades with somewhat shorterlatencies than targets of high spatial frequency (8–16 cpd).Earlier studies also reported longer manual reaction timesto foveally presented gratings of higher relative to lowerspatial frequency (Breitmeyer, 1975; Lupp, Hauske, &Wolf, 1976).

One diYculty in interpreting latency diVerences betweenstimuli of diVerent spatial frequency is that contrast sensi-tivity is not constant across spatial frequency and eccentric-ity (Rovamo et al., 1978; Wright & Johnston, 1983), andthat saccadic latency is highly sensitive to diVerences incontrast (Doma & Hallett, 1988). If stimuli are not carefullycontrolled in terms of detection performance, it is diYcultto rule out the contribution of contrast. Furthermore, atthreshold, saccade latencies are unavoidably noisy, so it isessential to perform a detailed analysis around detectionthreshold.

Only one study (Ludwig et al., 2004) examined the inXu-ence of spatial frequency on saccadic latency. In this studythe authors did not present targets speciWcally at threshold,but Wt a Piéron function (Pins & Bonnet, 1996) to eachobserver’s raw latency data plotted as a function of con-trast/contrast threshold (i.e., multiples of threshold). Whilethe higher SF targets showed some deviation (elevation)from the function for two observers, Ludwig and colleaguesnoted that the data for the third observer in fact convergedquite well into a single function (r2D0.79 to 0.86 for indi-vidual target locations). Furthermore, it is quite remarkablethat little diVerence was seen for spatial frequencies below8 cpd for either observer. This suggests that, to a largedegree, the spatio-temporal tuning for perception is thesame for the mechanisms driving saccadic eye movements.

The aim of the current study was to test whether themechanisms driving saccades do rely on the same spatio-temporal signals as perception, by (1) performing a moredetailed examination around detection threshold, (2) byusing a greater number of observers than in the Ludwiget al. (2004) study, and (3) by examining targets along thetemporal frequency (TF) domain as well, since both spatialand temporal frequency selectivity are important attributesof primate and human visual systems (De Valois et al.,1982; Derrington & Lennie, 1984; Foster, Gaska, Nagler, &

Pollen, 1985). Furthermore, the transient versus sustainednature of the magno- versus parvo-cellular pathways hasbeen a common explanation for diVerences in responselatency (Breitmeyer, 1975; Legge, 1978; Murray & Plainis,2003; Schwartz, 1992). Neurons in the magno-cellularlayers of the LGN respond better to rapidly Xickering lowspatial frequency targets, whereas neurons in parvo-cellularlayers can respond better to low temporal frequencies, andcan resolve much higher spatial frequencies (Derrington &Lennie, 1984; Merigan & Maunsell, 1993). It was thereforein our interest to investigate both spatial and temporaltarget properties.

Thus, our predictions are as follows: If the mechanismsdriving saccades do rely on the same spatio-temporalsignals as perception, the pattern of saccadic latenciesshould reXect perceptual sensitivity. That is, at equal detec-tion performance, latencies should be the same regardlessof spatio-temporal frequency. Alternatively, if the systemsrely on the output of diVerent visual mechanisms, weshould observe some systematic variation in latency as afunction of spatio-temporal frequency. In addition, wemeasured saccadic position error. While position error isbelieved to be independent of latency (Becker & Jurgens,1979; Findlay & Walker, 1999), a tradeoV may occur whentargets are more diYcult to detect (CoeVe & O’Regan,1987), making it pertinent to the current study. We primar-ily wanted to ensure that accuracy was held constant whenmaking a comparison between latencies. Finally, we alsocomputed oculometric- and psychometric-performance(proportion correct saccade direction versus psychophysi-cal detection) at threshold as a means of comparing thedegree to which the systems use the same visual spatio-tem-poral signals for decision processes. A similar method hasbeen used to compare saccadic- and perceptual-perfor-mance during visual search (Beutter, Eckstein, & Stone,2003), the results of which support the idea that both sys-tems use similar visual processing mechanisms for detec-tion. If saccadic and perceptual decisions rely on the samespatio-temporal signals, oculometric and psychometricperformance at threshold should not diVer.

2. Methods

2.1. Observers

One of the authors plus Wve additional naïve observers took part inExperiment 1, and three additional observers took part in Experiment 2.All observers had normal or corrected to normal visual acuity. Observersranged in age from 20 to 36 years of age. All observers received informedconsent before participating, and all experiments followed the ethicalguidelines of the Declaration of Helsinki.

2.2. Equipment

Stimuli were displayed on a 21 in. CRT monitor (ELO Touchsystems)driven by an ASUS V8170 GeForce 4 MX440 graphics board at a non-interlaced refresh rate of 100 Hz. The resolution of the monitor was set at1280 £ 1024 pixels, which corresponded to physical dimensions of 37 cmwide by 29.6 cm high. At a viewing distance of 47 cm, the display occupied

3888 B.J. White et al. / Vision Research 46 (2006) 3886–3897

a retinal area of 45 deg horizontally and 36 deg vertically. Eye-movementswere measured using EyeLink II (video-based tracker from SR ResearchLtd., Mississauga, Ontario) at a sample rate of 250 Hz.

2.3. Stimuli

The target was a vertically oriented Gabor patch (SD D 0.7 deg), pre-sented at one of four spatial frequencies (1, 2, 4 or 8 cpd), and oscillating atone of three temporal frequencies (1, 8 or 16 Hz). Targets appeared left orright of center at an eccentricity of 4 deg in Experiment 1, and 4 or 8 deg inExperiment 2. The central Wxation stimulus was a black spot approxi-mately 0.2 deg in diameter. Targets were presented on a neutral graybackground with a luminance of 32 cd/m2.

Because temporal frequency involves the modulation of contrastover time, targets of diVerent temporal frequencies may become visibleat diVerent times (e.g., a 16 Hz target will reach peak contrast soonerthan a 1 Hz target). This is an obvious diYculty when trying to comparelatencies between targets of diVerent temporal frequency. We dealt withthis in two ways: for the primary experiment (Experiment 1), targetswere onset at their maximum point of modulation for a given contrast,and the temporal oscillation continued from there for 500 ms. This pro-duced an abrupt onset for all targets, which could drive transient mecha-nisms (Legge, 1978; Murray & Plainis, 2003), and possibly overshadowsubtle diVerences from the target’s actual temporal characteristics. AFourier analysis revealed that only the 1 Hz target showed a signiWcantamount of residual energy, but it was spread out over a large range offrequencies, with most of the energy around 1–2 Hz. A second experi-ment (Experiment 2) was run where we ramped the onset of the 8 and16 Hz targets with the sinusoidal temporal contrast function of the 1 Hztarget. Here, target duration was 1000 ms to deal with the fact that tar-gets would become visible at a later time. At a monitor refresh of 100 Hz,the 1 Hz target lasted the duration of one complete cycle, and reachedmaximum contrast after 25 frames (250 ms), or one quarter of the entirecycle. The Wrst quarter of the 8 and 16 Hz targets was multiplied with theWrst quarter of the 1 Hz sinusoid. So in essence, the 8 and 16 Hz targetsretained their primary Xicker frequency throughout the entire period.The only diVerence was their onset was essentially windowed by the 1 Hzcontrast modulation. In this way, the time to reach maximum contrastwas the same for all targets, 250 ms after initial onset. Analysis of theFourier energy along the temporal domain showed negligible 1 Hz resid-ual energy for the 8 and 16 Hz targets.

Typically saccade targets are presented with an abrupt onset. Giventhat response latencies have been attributed to the transient versus sus-tained nature of the magno- versus parvo-cellular pathways (Breitmeyer,1975; Legge, 1978; Murray & Plainis, 2003; Schwartz, 1992), it was in ourinterest to perform tests using both abrupt and slower onset targets.

2.4. Procedure

2.4.1. ThresholdsWe used a staircase procedure to determine contrast threshold for each

of the 12 targets (4 SFs £ 3 TFs) in a 2-alternative forced choice (2AFC)procedure for Experiment 1 (left versus right), and a 4AFC in Experiment2 (left-near, left-far, right-near, and right-far). Two staircases per stimuluswere initiated, one at a log unit below the starting value, and one at a logunit above. Starting values were set relatively low such that stimuli werejust visible. The staircase followed a 3-up 1-down rule yielding the 79%detection threshold (see Macmillan & Creelman, 1991): three correctresponses in a row resulted in a contrast decrease of 1 log unit, and oneincorrect response resulted in a contrast increase of 1 log unit. The thresh-old of each stimulus was the mean of eight reversal points (four for eachstaircase).

Note that eye movements were controlled using the Eyelink II tracker,and the head was stabilized by a chin rest. Calibrations were made beforeeach block of trials (approximately every 50–100 trials), and consisted ofWxating 9 consecutive bull’s-eye stimuli at various locations on the screen.Average spatial accuracy for each calibration was maintained at 0.35 degor better.

The Wxation stimulus was present before the trial was initiated. Observ-ers had to Wxate this stimulus, and then initiate the trial when they wereready by pressing a key on a game-pad. This allowed for a drift correctionprocedure at the start of each trial. In the event that observers were notWxating accurately (within 1 deg), an error-tone was presented, and thetrial had to be reinitiated. If a saccade of greater than 1 degree was madeduring a trial, a visual and auditory warning was presented, and the trialwas aborted with no change in the status of the staircase. This ensured thatthe resulting thresholds were not confounded by periodically Wxating thetarget. This was aided by the presence of the Wxation stimulus throughoutthe duration of a trial. The target was onset randomly between 800 and1200 ms after a trial was initiated, and appeared left or right of Wxation atan eccentricity of 4 deg in Experiment 1, and 4 or 8 deg in Experiment 2.The target remained present for 500 ms in Experiment 1, and 1000 ms inExperiment 2. All trials were randomly interleaved. Observers simplymade a 2AFC (Experiment 1, left, right) or a 4AFC (Experiment 2, left-near, left-far, right-near, and right-far) as to the location of the target.

2.4.2. Eye movementsOnce we determined thresholds, each stimulus was presented at various

multiples of this during a separate session (0.8, 1, 2, 3, 4 or 10£ thresholdand 100% contrast in Experiment 1, and 1, 1.1, 2, 4£ threshold and 100% inExperiment 2). Note that due to the inherent noise at threshold, observationsfrom the two lowest contrasts (0.8 and 1£ threshold for Experiment 1, and 1and 1.1£ thresh. for Experiment 2) were later pooled to increase the reliabil-ity of the mean at this level of visibility.

Observers were simply requested to make an eye movement to the cen-ter of the target when it appeared, and that both speed and accuracy areequally important. The stimulus presentation was identical to the thresh-old procedure above except that the Wxation stimulus was removed at thesame time the target was onset. As with the threshold procedure, after eachtrial, observers made a psychometric decision as to the location of the tar-get by pressing the appropriate key. Eye position and event data wererecorded, and all analyses were done oZine.

2.5. Analysis

Saccadic latency was used as the primary indicator of the sensitivity ofthe saccadic system to diVerent spatio-temporal frequencies. Latency was thetime (in ms) between target onset and the onset of the Wrst saccade. Saccadeswere detected by the EyeLink-II system at a velocity threshold of 30 deg/s oran acceleration threshold of 8000 deg/s2. We also computed position errorwhich was the Euclidean distance (unsigned) between the end point of theWrst saccade and the target center in degrees. We primarily wanted to ensurethat position error was held constant when making a comparison betweenlatencies. Furthermore, at low contrasts, if position error is large, it suggeststhat the saccade may not have been driven by the target (but might insteadbe a voluntary guess). We tried to rule out the contribution of such trials bycomputing latencies only for trials where position error was less than 2 deg.The same criterion was used by Ludwig and colleagues (2004). Finally, wealso measured the proportion of correct saccade directions (oculometric per-formance) at psychophysical detection threshold to determine the degree towhich saccadic and perceptual decisions rely on the same visual signals (i.e.,with regard to spatial and temporal frequency). As mentioned earlier, Beut-ter and colleagues (2003) used a similar method to compare saccadic- andperceptual performance during visual search, and argued that both systemsuse similar mechanisms for visual detection. For our purposes, this methodsimply provided and additional dimension for comparing both systems.

Observers completed approximately 1500 trials each in Experiment 1,and 1000 trials each in Experiment 2, with the exception of observers LPand NB in Experiment 1, and observer SO in Experiment 2 who completedjust over 500 trials each. No outlier procedure was used, but we consideredsaccades with latencies less than 80 ms as anticipatory responses (see Wen-ban-Smith & Findlay, 1991). Trials with saccadic direction errors werealso removed from the accuracy and latency analyses. With errorsremoved, the mean of each target condition consisted of 10–30 observa-tions per observer (with the exception of observers LP, NB and SO whohad somewhat fewer on some target conditions).

B.J. White et al. / Vision Research 46 (2006) 3886–3897 3889

3. Experiment 1

3.1. Results and discussion

3.1.1. Proportion of errorsTable 1 shows the proportion of saccades and error tri-

als at detection threshold. For brevity, the table showsthreshold trials only since this is where most errorsoccurred. At detection threshold, the proportion of trials inwhich observers did not make a saccade was less than 25%,with the exception of observers MS and NB. This does notnecessarily mean that detection performance was worse forthese observers, but that their criterion for making a sacc-adic response may have been rather high. Note that theproportion of saccade direction errors (saccades in oppositedirection to target) was computed from only those trialsin which a saccade occurred because the number ofdirection errors out of the total trials would overestimateperformance for those who made less saccades (especiallyobservers MS and NB). To make detection performancecomparable to this, it was also computed from saccade tri-als only. As such, direction and detection performance werequite similar (as seen in Table 1), and close to the predictedthreshold performance based on the staircase rule (79%, seeSection 2). Furthermore, a correlation between saccadedirection errors and detection errors over trials was highlysigniWcant for each observer, �D 0.45 to 0.77, p < .001 in allcases. Note that except for observers NB and SW, the

proportion of inaccurate trials was low. Most removed tri-als were due to direction or detection errors at threshold.The last column represents the total proportion of saccadetrials used at threshold (errors and inaccurate trialsremoved).

Fig. 1 shows the mean proportion of saccadic directionversus detection errors at threshold, as a function of targettemporal frequency (left plot), and spatial frequency (rightplot). The mean was around 25–30% (i.e., 70–75% correct)across all temporal and spatial frequencies, with only slightlyhigher direction versus detection errors overall, which maybe due to the fact that saccades necessarily involve a speededresponse, possibly making them more prone to directionerrors. A 2£3 (error-type£ temporal frequency) and 2£4(error-type£ spatial frequency) repeated measures ANOVAdid reveal a small eVect of error-type only (F (1,5)D9.8,p < .05, and F(1,5)D9.7, p < .05, respectively). However, F wasless than 1 for all remaining eVects, which means that oculo-metric- and psychometric-performance did not diVer acrossspatial or temporal frequency. This supports the idea thatsaccadic and perceptual decisions rely on the same visualspatio-temporal signals.

3.1.2. AccuracyAs mentioned earlier, an accuracy-latency tradeoV may

occur when targets are diYcult to detect (CoeVe &O’Regan, 1987). This analysis was performed to ensurethere was no such tradeoV. A 2£ 3£ 4 (2 contrasts£ 3

Table 1Proportion of saccades, errors and used trials at detection threshold (Experiment 1)

The proportion of direction and detection errors was computed from trials with saccades only (Wrst column). The proportion correct represents thesaccade trials with no direction or detection errors, and the proportion used represents the correct trials (fourth column) that satisWed the inaccuracy/anticipatory criterion (see Section 2).

Observer Proportion saccades Direction errors Detection errors Proportion correct Inaccurate/anticipatory Proportion used (N)

BW 0.89 0.34 0.24 0.64 0.04 0.61 (322)LP 0.77 0.30 0.28 0.69 0.13 0.60 (87)MS 0.57 0.29 0.23 0.67 0.06 0.63 (188)NB 0.39 0.23 0.22 0.76 0.29 0.54 (54)NZ 0.85 0.26 0.22 0.74 0.16 0.62 (188)SW 0.76 0.12 0.09 0.88 0.24 0.67 (167)

Fig. 1. Mean proportion of saccade direction errors and detection errors at threshold as a function of temporal frequency (left) and spatial frequency(right) (Experiment 1).

3890 B.J. White et al. / Vision Research 46 (2006) 3886–3897

TFs£ 4 SFs) repeated measures ANOVA was run on thedata to test whether any diVerences exist between spatialor temporal frequency. We chose only the two lowest con-trasts (threshold and twice threshold) because multiples ofthreshold greater than this were not possible for allobservers on all targets, especially in terms of the highspatial frequencies.1 If diVerences do exist, they should bemore easily detected near threshold. The ANOVArevealed a small eVect of SF (F (3, 15)D 4.6, pD .017) and aSF£ contrast interaction (F (3, 15)D 4.4, pD .021). Noother eVects were signiWcant (p > 0.3 for all remainingtests). The eVect of SF was due to slightly higher error for

1 Because of limitations in the maximum contrast possible on CRTmonitors, contrasts of more than 2£ threshold were not always possiblefor all observers on all targets, especially in terms of the high spatial fre-quencies.

1 cpd targets (0.51 deg) relative to 2 cpd targets (0.44 deg),p < .05, but no other diVerences were signiWcant with aBonferroni correction. In short, there was little diVerencein saccadic error across targets of diVerent spatio-tempo-ral frequency.

3.1.3. LatencyFig. 2 shows saccadic latency as a function of contrast

for three representative observers. Fig. 2A is plotted forcomparison between spatial frequencies within a given tem-poral frequency, and Fig. 2B is for comparison betweentemporal frequencies within a given spatial frequency. TheWrst thing to note is the predicted steady decrease in latencyas a function of contrast. Second, consistent with the Lud-wig et al. (2004) study, at a given contrast, latencies werelonger at consecutively higher spatial frequencies (Fig. 2A).This was similar for the highest temporal frequency

Fig. 2. Saccadic latency as a function contrast (log) for three representative observers (Experiment 1), plotted for comparison between spatial frequencieswithin a given temporal frequency (A), and between temporal frequencies within a given spatial frequency (B). Errorbars omitted for clarity.

B.J. White et al. / Vision Research 46 (2006) 3886–3897 3891

(Fig. 2B), but overall there was considerably more overlapbetween temporal frequencies. The results were similar forthe other observers.

Fig. 3 shows the mean saccade latency of the sixobservers, plotted for comparison between spatial fre-quencies (top), and between temporal frequencies (bot-tom). As can be seen, latencies showed a steady decreaseas a function of multiples of contrast threshold, but diVer-ences between spatial and temporal frequencies weremuch less evident here.

We ran a 2£3£4 (2 contrasts£3 TFs£4 SFs) repeatedmeasures ANOVA on the latency data. As with saccadicerror, we chose only the two lowest contrasts (threshold andtwice threshold) for the ANOVA because multiples ofthreshold greater than this were not possible for all observerson all targets. Furthermore, if any diVerences exist, theyshould be most detectable near threshold. The ANOVArevealed a strong eVect of contrast (F (1,5)D54, pD .001),and a TF£SF interaction (F (6,30)D3.1, pD .017). No othereVects were statistically signiWcant. The eVect of contrast wasreasonably clear from the plots, but the interaction was diY-

cult to interpret. Table 2 shows the mean latencies (plus SEs)across SF for each TF. We ran a repeated measures ANOVAacross SF for each TF, and across TF for each SF while cor-recting for multiple comparisons. The only signiWcant diVer-ence found was across TF for the 2 cpd targets (pD .021 withBonferroni correction), with slightly longer latencies athigher TFs (283, 315, and 347ms, for the 1, 8, and 16 Hz tar-gets, respectively). No other signiWcant diVerences werefound (p > .25 for all remaining tests).

Thus, at equal detection performance, the results showedlittle systematic diVerence in saccade latency as a function ofspatial or temporal frequency. While the main eVect of tem-poral frequency was consistent in one direction (longer laten-

cies at successively higher TFs), it was true only for the 2cpdtargets, and no other such consistencies were observed. Thelack of an eVect of SF here disagrees with the results of theLudwig et al. (2004) study. In fact, for our latency data, F wasless than 1 for both the main eVect of SF and theSF£contrast interaction. It is possible that diVerences existat SFs greater than 8cpd, but the eVect appeared about thesame for 8 or 16cpd targets in the Ludwig et al. study. Moreimportantly, our data showed no indication of a systematicordering of latencies from low to high spatial frequency aswas reported in the Ludwig et al. (2004) study. This would beimportant for the idea that latency increases as a function ofspatial frequency. This was not the case here. In short, theresults of Experiment 1 suggest that, to a large degree, themechanisms controlling saccadic eye movements have accessto the same spatio-temporal signals used for perception. It ispossible however that the lack of a consistent eVect of TFwas due to the transient target onset across all conditions.Experiment 2 was designed to test this possibility.

4. Experiment 2

The abrupt onset of the target in Experiment 1 didproduce residual high temporal frequency energy, espe-cially in terms of the 1 Hz targets. This may have masked

Table 2Mean saccade latency and standard error (in parentheses) for the TF £ SFinteraction (Experiment 1)

TF SF

1 cpd 2 cpd 4 cpd 8 cpd

1 Hz 315 ms (30) 283 ms (23) 322 ms (29) 313 ms (21)8 Hz 343 ms (28) 315 ms (27) 326 ms (22) 332 ms (20)16 Hz 334 ms (23) 347 ms (28) 323 ms (29) 313 ms (29)

Fig. 3. Mean saccadic latency (6 observers) as a function of multiples of contrast threshold (Experiment 1), plotted for comparison between spatial fre-quencies within a given temporal frequency (top), and between temporal frequencies within a given spatial frequency (bottom). Errorbars represent §1standard error.

3892 B.J. White et al. / Vision Research 46 (2006) 3886–3897

any diVerences due to the target’s actual temporal fre-quency characteristics. Using an abrupt onset was the sim-plest way of dealing with the fact that targets of diVerenttemporal frequencies reach peak contrast at diVerent times,making latency comparisons diYcult. Here, we ramped theonset of the 8 and 16 Hz targets with the sinusoidal contrastfunction of the 1 Hz target (see Section 2). In addition, weused two target eccentricities (4 or 8 deg) to reduce the pre-dictability of the target’s position.

4.1. Results and discussion

4.1.1. Proportion of errorsTable 3 shows the proportion of saccades and error tri-

als at detection threshold for Experiment 2. For brevity, thetable shows threshold trials only since this is where mosterrors occurred. At detection threshold, the proportion ofno-saccade trials was less than 10%. As with Experiment 1,direction and detection performance were similar, butslightly higher then the predicted threshold performancebased on the staircase rule (79%, see Section 2). This may bedue to the extended target duration, or the fact that wepooled data from threshold and 1.1£ threshold in this case.The proportion of inaccurate trials was on average slightlyhigher here (MD 0.18) than in Experiment 1 (MD0.15).

Fig. 4 shows the mean proportion of saccadic directionversus detection errors at threshold, as a function of targettemporal frequency (left plot), and spatial frequency (rightplot). The mean was around 10–15% (i.e., 85–90% correct)

across temporal and spatial frequencies. A 2£ 3 (error-type£ temporal frequency) and 2£ 4 (error-type£ spatialfrequency) repeated measures ANOVA revealed no signiW-cant diVerences (p > .25 for all eVects). The lack of an eVectof error-type may be due to the fact that saccadic perfor-mance was necessarily computed as a binary response (leftversus right), whereas psychophysical performance was a4AFC procedure (left-near, left-far, right-near, and right-far). Nonetheless, most importantly there was no indicationof a diVerence across spatial or temporal frequency withineither response-type. In other words, there was no biastowards a particular spatial or temporal frequency. Thisagain supports our claim that the saccadic and perceptualsystems rely on the same spatio-temporal signals.

4.1.2. AccuracyAs with Experiment 1, this analysis was performed to

ensure there was no accuracy-latency tradeoV. We ran testsseparately for each target eccentricity to include as manythreshold multiples as possible. A 2£ 3£ 4 (2 contrasts£ 3TFs£ 4 SFs) repeated measures ANOVA was run on thetargets at 4 deg eccentricity. The ANOVA revealed an eVectof contrast only (F (1, 3)D26.1, pD .015), with slightlyhigher saccadic error at threshold (0.7 deg) than at thehigher contrast (0.59 deg). No other eVects at this eccentric-ity were signiWcant (p > .15 for all remaining tests). For theoutermost targets, it was only possible to test all SFs andTFs at detection threshold (see Footnote 1). However, noneof the eVects were signiWcant (F < 1 for the main eVect of

Table 3Proportion of saccades, errors and used trials at detection threshold (Experiment 2)

The proportion of direction and detection errors was computed from trials with saccades only (Wrst column). The proportion correct represents the sac-cade trials with no direction or detection errors, and the proportion used represents the correct trials (fourth column) that satisWed the accuracy/anticipa-tory criterion (see Section 2).

Observer Proportion saccades Direction errors Detection errors Proportion correct Inaccurate/anticipatory Proportion used (N)

BW 0.97 0.17 0.15 0.79 0.16 0.66 (247)DW 0.93 0.06 0.04 0.92 0.14 0.79 (284)PS 0.91 0.08 0.08 0.89 0.19 0.72 (255)SO 0.98 0.08 0.18 0.81 0.24 0.62 (117)

Fig. 4. Mean proportion of saccade direction errors and detection errors at threshold as a function of temporal frequency (left) and spatial frequency(right) (Experiment 2).

B.J. White et al. / Vision Research 46 (2006) 3886–3897 3893

TF; F (3, 9)D1.2, pD .34 for the main eVect of SF;F (6, 18)D 1.1, pD .38 for the TF£SF interaction). We wereable to run another test including two threshold multiples,but only with 2 TFs and 4 SFs (2£ 2£4). Again, only theeVect of contrast was signiWcant (F (1,3)D 13.5, pD .035),showing slightly higher saccadic error at threshold(0.89 deg) than at the higher contrast (0.78 deg). In mostcases, F was less than 1 (e.g., for the main eVect of TF, SF,

TF£SF, TF£ contrast, and TF£SF£ contrast). Thus, aswith Experiment 1, position error remained fairly constantacross target spatio-temporal frequency.

4.1.3. LatencyFig. 5 shows saccadic latency as a function of contrast

for two representative observers. Fig. 5A is plotted for com-parison between SFs within a given TF, and Fig. 5B for

Fig. 5. Saccadic latency as a function contrast (log) for two representative observers (Experiment 2), at two target eccentricities (4 or 8 deg), plotted forcomparison between spatial frequencies within a given temporal frequency (A), and between temporal frequencies within a given spatial frequency (B).Errorbars omitted for clarity.

3894 B.J. White et al. / Vision Research 46 (2006) 3886–3897

comparison between TFs within a given SF. Latencies wereoverall much longer here due to the ramped target onset(see Section 2), but the pattern was similar to Experiment 1in that we see a steady decrease in latency as a function ofcontrast. The pattern in terms of TF (Fig. 5B) was some-what diVerent from before: Fig. 5B shows considerableoverlap between 1 and 8 Hz targets at lower SFs (which issimilar to the pattern in Experiment 1; see Fig. 2B), butthere is an increasing separation between the TFs atincreasing SFs in Experiment 2 (Fig. 5B) that is not so obvi-ous in Experiment 1 (Fig. 2B). If we consider a Wxed con-trast at low TF (Fig. 5A), latencies are somewhat longer forhigher SF targets, but this diVerence appears to increase athigher TFs, and with greater target eccentricity (the linesseparate more than in Fig. 2A, Experiment 1). This is in factwhat we would expect based on the spatio-temporal con-trast sensitivity function (Kelly, 1984): spatial contrast sen-

sitivity tends to be band-pass at low TFs and low-pass athigh TFs. This means that we are better able to resolve fastXickering low SF patterns then slow Xickering high SF pat-terns, and this diVerence should be more pronounced fur-ther in the periphery due its greater sensitivity to Xicker(Baker & Braddick, 1985), and poorer sensitivity to highSF. The data in Experiment 2 are consistent with this.

However, the question is to what extent this diVerenceholds at points of equal detection performance. Fig. 6shows the mean saccadic latency across threshold multiplesfor the four observers, plotted for comparison between SFs(top two), and between TFs (bottom two), at each targeteccentricity. As with accuracy, latencies were noisier athigher threshold multiples, especially for the higher SFs(4 and 8 cpd) at the outermost eccentricity, because some ofthese points represent data from as few as one observer

Fig. 6. Mean saccade latency (4 observers) as a function of multiples of contrast threshold (Experiment 2), plotted for comparison between spatial frequen-cies within a given temporal frequency for each target eccentricity (top two sets of panels), and between temporal frequencies within a given spatial fre-quency (bottom two sets of panels). Errorbars represent §1 standard error.

B.J. White et al. / Vision Research 46 (2006) 3886–3897 3895

(Footnote 1). However, at equal detection performance (i.e.,near threshold), there appears to be little if any diVerence.

We ran tests separately for each target eccentricity inorder to include as many threshold multiples as possiblewhile ensuring that each condition contained data from allfour observers (since most missing cases were for targets atthe outermost eccentricity at the highest SF; Footnote 1).We were able to run a 2£3£ 4 (2 contrasts£ 3 TFs£4SFs) repeated measures ANOVA on the targets at 4 degeccentricity. The ANOVA revealed a main eVect of SF(F (3, 9)D 6.9, pD .01), and contrast (F (1, 3)D397, p < .001),plus a TF£ contrast interaction (F (2, 6)D 5.4, pD .04), anda SF£ contrast interaction (F (3,9)D 5.4, pD .02). No othereVects were signiWcant.

The main eVect of contrast showed an expected longerlatency at threshold (622 ms) than twice threshold(445 ms). However, for the SF main eVect, Bonferroniadjusted comparisons between the speciWc SFs (537, 521,562, and 513 ms for the 1, 2, 4, and 8 cpd targets respec-tively) did not reach statistical signiWcance (p > .09 for allcomparisons). In addition, the interactions showed noobvious systematic diVerences. Table 4 shows the meansaccade latency (and SEs) for both interactions thatreached statistical signiWcance. We ran a repeated mea-sures ANOVA at both contrasts (1 and 2 £ thresh) acrossTF (Table 4A), and then across SF (Table 4B), while cor-recting for multiple comparisons. Only the eVect of SF atthreshold was signiWcant (F (3, 9)D 9.1, p < .05), and was

Table 4Mean saccade latency and standard error (in parentheses) for theTF £ contrast interaction (A) and the SF £ contrast interaction (B) forExperiment 2

(A) TF

Contrast 1 Hz 8 Hz 16 Hz

Threshold 604 ms (50) 632 ms (32) 629 ms (25)2£ Threshold 446 ms (37) 452 ms (30) 437 ms (21)

(B) SF

Contrast 1 cpd 2 cpd 4 cpd 8 cpd

Threshold 624 ms (38) 626 ms (44) 662 ms (33) 574 ms (25)2£ Threshold 449 ms (30) 417 ms (23) 464 ms (28) 452 ms (37)

due to slightly longer latency for 4 cpd targets (662 ms)relative to 8 cpd targets (574 ms). Note that this is in factin the opposite direction to the eVect reported by Ludwigand colleagues (2004). In their study, latencies were in factelevated for SFs >D 8 cpd.

For targets at 8 deg eccentricity, it was only possible totest all SFs and TFs at detection threshold (multiplesgreater than this were not possible for all targets; see Foot-note 1). However, none of the eVects were signiWcant (p > .15for all eVects). We were able to run another test includingtwo threshold multiples, but with only 2 TFs and 4 SFs(2£ 2£4). Again, there was an expected eVect of contrast(F (1, 3)D231, p < .001). The only other eVect to reach statis-tical signiWcance was TF (F (1,3)D 12.7, pD .04), and wasdue to slightly longer latencies for 8 Hz (557 ms) than 1 Hz(516 ms) targets.

To reinforce these results, a more stringent test was runat threshold with a single observer using only two SFs (1and 8 cpd), two TFs (1 and 16 Hz), and one target eccentric-ity (8 deg). For the staircase, threshold was taken as themean of 8 reversals (versus 4 previously). We obtained 100trials per stimulus condition at only one contrast, detectionthreshold. If saccadic latency is in fact longer for high SFs(Ludwig et al., 2004), it should be detectable here. Fig. 7shows the results. On the left is saccade latency, and on theright is saccade error, with the percentage of correct sac-cade directions next to each mean value. Again, there wasno indication of longer latencies for high SF targets here.

5. General discussion

We can conclude from these experiments that perceptualand saccadic performance was remarkably similar: oncetargets were presented at near equal detection performance,we found little systematic diVerence in saccade latency andno diVerence in average oculometric performance (% cor-rect saccade direction) across a range of spatial and tempo-ral frequencies. Furthermore, saccadic position errorremained fairly constant across all conditions. The onlyreliable eVect was due to contrast. The pattern of latenciesas a function of contrast (see Figs. 2 and 5) was what wewould expect based on the spatio-temporal contrast sensi-

Fig. 7. Saccade latency (left) and saccade error (right) as a function of target spatio-temporal frequency (targets at detection threshold). Errorbarsrepresent §1 standard error. Percent correct saccade direction plotted to the side.

3896 B.J. White et al. / Vision Research 46 (2006) 3886–3897

tivity function (Kelly, 1984). In short, sensitivity falls athigher SFs, which is reXected in the pattern of latencies as afunction of contrast (Figs. 2 and 5) but this diVerencebegins to disappear when we control for contrast sensitivityby performing tests at multiples of threshold (Figs. 3 and 6).Taken together, the results support the idea that themechanisms driving saccades rely on the same visualspatio-temporal signals as perception.

It is possible that latency diVerences exist at spatial fre-quencies higher than was used in the current study (8 cpd).The Ludwig et al. (2004) study used spatial frequencies up to16 cpd, and they reported a systematic ordering of latenciesfrom medium to high spatial frequencies. However, oursetup did not permit an adequate sampling of 16 cpd targets.By our accounts, this was also a problem in the Ludwig et al.study (approximately 2 pixels per cycle at 16 cpd, which isfairly close to a square-wave pattern). Furthermore, we didnot Wnd any indication of a systematic ordering of latenciesfor spatial frequencies up to 8 cpd, so it seems doubtful thatwe would have found a diVerence at 16 cpd. However, fur-ther tests using a more detailed sampling resolution for spa-tial frequencies beyond 8 cpd might be useful.

The use of temporal frequency was also an importantaspect of this study, and arguably it has been understudiedin terms of saccades. It has been suggested that diVerencesin response latencies due to stimuli with diVerent spatio-temporal properties can be attributed to the transientversus sustained nature of the magno- versus parvo-cellularpathways, respectively (Breitmeyer, 1975; Legge, 1978;Murray & Plainis, 2003; Schwartz, 1992). Neurons in themagno-cellular layers of the LGN respond better to rapidlyXickering low spatial frequency targets, whereas neurons inparvo-cellular layers can respond better to low temporalfrequencies, and can resolve much higher spatial frequen-cies (Derrington & Lennie, 1984; Merigan & Maunsell,1993). However, the conduction time from the optic chiasmto the LGN is only about 3–4 ms slower for the P-layersrelative to the M-layers (Schiller & Malpeli, 1978), and thediVerence is only slightly larger in terms of visual responselatency, with reports from 7–10 ms (Maunsell et al., 1999;Maunsell & Gibson, 1992). For behavioral diVerences tomatch this would require that visual stimuli uniquely iso-late each pathway. It seems unlikely then that stimulibelieved to isolate these channels would produce a behav-iorally signiWcant diVerence. We used both abrupt-onsettargets (Experiment 1) and slow-onset targets (Experiment2) of diVerent temporal frequencies, but the results showedno clear consistent eVect to support this idea. In fact, atequal detection performance, there was no diVerence acrosstemporal frequency for every spatial frequency except2 cpd, where the eVect was actually opposite to the predic-tion based on the fast M-path versus the slower P-path (i.e.,latencies were slightly longer for successively higher TFs).

One might argue that the few diVerences we did Wnd (orarguably the lack of any diVerences) were due to less thanperfect thresholds. The staircase method is a quick meansof obtaining the threshold for many stimuli simultaneously,

but plotting a psychometric function may be a more accu-rate approach. Unfortunately, it was not possible to ade-quately Wt a psychometric function to our data (which isoften the case for data obtained from a staircase). The alter-native is to use a constant stimuli procedure, but thetradeoV is time due to the greater amount of data requiredto Wt a reasonable function. This might be worth exploringwith fewer target spatio-temporal frequencies.

In contrast to our results, there has been considerable sup-port for the idea that the visual signals driving perception aredistinct from the visual signals used to guide actions (e.g.,Goodale & Milner, 1992; Goodale & Westwood, 2004). Partof the support comes from patients with lesions in areasbelieved to be primarily responsible for motor behavior (so-called “dorsal” areas) versus visual perception (so-called“ventral” areas). For example, patients with lesions in dorsalareas have shown deWcits in reaching and grasping certainobjects, while their perception of such objects is presumablyunaVected (Perenin & Vighetto, 1988). Conversely, patientswith ventral lesions have shown perceptual deWcits whilemotor behavior is presumably spared (so-called visual formagnosia; Milner et al., 1991). Since this initial hypothesishowever, there has been growing evidence suggesting a highdegree of interaction between dorsal and ventral areas, and amuch higher level visual representation in the dorsal stream.For example, several studies have reported eVects of visualillusions on reaching and grasping behavior (e.g., de Grave,Brenner, & Smeets, 2004; Franz, Gegenfurtner, BulthoV, &Fahle, 2000; Glover & Dixon, 2001, 2002). In addition, stud-ies are beginning to suggest that the motor system can usepure chromatic signals to guide rapid pointing behavior(Brenner & Smeets, 2004 ;Schmidt, 2002). Furthermore, thereis some evidence for color selectivity in dorsal area LIP (lat-eral intraparietal area) when color is relevant for the task, inthis case a saccade task (Toth & Assad, 2002). Along theselines, the results of the current study support the idea thatdorsal areas in the control of saccadic eye movements haveaccess to the same spatio-temporal signals as perception.

Acknowledgments

This research was funded by the Bundesministerium fürBildung und Forschung (“Modkog” 62000177), and theDFG Forschergruppe 560 (“Perception and Action”).

References

Baker, C. L., Jr., & Braddick, O. J. (1985). Eccentricity-dependent scalingof the limits for short-range apparent motion perception. VisionResearch, 25(6), 803–812.

Becker, W., & Jurgens, R. (1979). An analysis of the saccadic system bymeans of double step stimuli. Vision Research, 19(9), 967–983.

Beutter, B. R., Eckstein, M. P., & Stone, L. S. (2003). Saccadic and percep-tual performance in visual search tasks. I. Contrast detection and dis-crimination. Journal of the Optical Society of America A Optics, ImageScience, and Vision, 20(7), 1341–1355.

Breitmeyer, B. G. (1975). Simple reaction time as a measure of the tempo-ral response properties of transient and sustained channels. VisionResearch, 15(12), 1411–1412.

B.J. White et al. / Vision Research 46 (2006) 3886–3897 3897

Brenner, E., & Smeets, J. B. (2004). Colour vision can contribute to fastcorrections of arm movements. Experimental Brain Research, 158(3),302–307.

Bruce, C. J., & Goldberg, M. E. (1985). Primate frontal eye Welds I: singleneurons discharging before saccades. Journal of Neurophysiology, 53,603–635.

Carpenter, R. H. S. (1988). Movements of the eyes. London: Pion.CoeVe, C., & O’Regan, J. K. (1987). Reducing the inXuence of non-target

stimuli on saccade accuracy: predictability and latency eVects. VisionResearch, 27(2), 227–240.

de Grave, D. D., Brenner, E., & Smeets, J. B. (2004). Illusions as a tool tostudy the coding of pointing movements. Experimental Brain Research,155(1), 56–62.

De Valois, R. L., Albrecht, D. G., & Thorell, L. G. (1982). Spatial fre-quency selectivity of cells in macaque visual cortex. Vision Research,22(5), 545–559.

Derrington, A. M., & Lennie, P. (1984). Spatial and temporal contrast sen-sitivities of neurones in lateral geniculate nucleus of macaque. Journalof Physiology, 357, 219–240.

Doma, H., & Hallett, P. E. (1988). Rod-cone dependence of saccadic eye-movement latency in a foveating task. Vision Research, 28(8), 899–913.

Findlay, J. M., & Gilchrist, I. D. (2003). Active vision: The psychology oflooking and seeing. Oxford University Press.

Findlay, J. M., & Walker, R. (1999). A model of saccade generation basedon parallel processing and competitive inhibition. Behavioural andBrain Sciences, 22(4), 661–674 discussion 674-721.

Fischer, B., & Ramsperger, E. (1984). Human express saccades: extremelyshort reaction times of goal directed eye movements. ExperimentalBrain Research, 57(1), 191–195.

Foster, K. H., Gaska, J. P., Nagler, M., & Pollen, D. A. (1985). Spatial andtemporal frequency selectivity of neurones in visual cortical areas V1and V2 of the macaque monkey. Journal of Physiology, 365, 331–363.

Franz, V. H., Gegenfurtner, K. R., BulthoV, H. H., & Fahle, M. (2000).Grasping visual illusions: no evidence for a dissociation between per-ception and action. Psychological Science, 11(1), 20–25.

Glover, S., & Dixon, P. (2001). The role of vision in the on-line correctionof illusion eVects on action. Canadian Journal of Experimental Psychol-ogy, 55(2), 96–103.

Glover, S., & Dixon, P. (2002). Dynamic eVects of the Ebbinghaus illusionin grasping: support for a planning/control model of action. Perception& Psychophysics, 64(2), 266–278.

Goodale, M. A., & Milner, A. D. (1992). Separate visual pathways for per-ception and action. Trends in Neurosciences, 15(1), 20–25.

Goodale, M. A., & Westwood, D. A. (2004). An evolving view of duplexvision: separate but interacting cortical pathways for perception andaction. Current Opinion in Neurobiology, 14(2), 203–211.

Kalesnykas, R. P., & Hallett, P. E. (1994). Retinal eccentricity and thelatency of eye saccades. Vision Research, 34(4), 517–531.

Kelly, D. H. (1979). Motion and vision. II. Stabilized spatio-temporalthreshold surface. Journal of the Optical Society of America, 69(10),1340–1349.

Kelly, D. H. (1984). Retinal inhomogeneity. I. Spatiotemporal contrastsensitivity. Journal of the Optical Society of America A, 1(1), 107–113.

Legge, G. E. (1978). Sustained and transient mechanisms in human vision:temporal and spatial properties. Vision Res., 18(1), 69–81.

Ludwig, C. J., & Gilchrist, I. D. (2002). Stimulus-driven and goal-drivencontrol over visual selection. Journal of Experimental Psychology:Human Perception and Performance, 28(4), 902–912.

Ludwig, C. J., Gilchrist, I. D., & McSorley, E. (2004). The inXuence of spa-tial frequency and contrast on saccade latencies. Vision Research,44(22), 2597–2604.

Lupp, U., Hauske, G., & Wolf, W. (1976). Perceptual latencies to sinusoi-dal gratings. Vision Research, 16(9), 969–972.

Macmillan, N. A., & Creelman, C. D. (1991). Detection theory: Auser’sguide. New York: Cambridge University Press.

Maunsell, J. H., Ghose, G. M., Assad, J. A., McAdams, C. J., Boudreau, C.E., & Noerager, B. D. (1999). Visual response latencies of magnocellu-lar and parvocellular LGN neurons in macaque monkeys. Visual Neu-roscience, 16(1), 1–14.

Maunsell, J. H., & Gibson, J. R. (1992). Visual response latencies in striatecortex of the macaque monkey. Journal of Neurophysiology, 68(4),1332–1344.

Merigan, W. H., & Maunsell, J. H. (1993). How parallel are the primatevisual pathways? Annual Review of Neuroscience, 16, 369–402.

Milner, A. D., Perrett, D. I., Johnston, R. S., Benson, P. J., Jordan, T. R.,Heeley, D. W., et al. (1991). Perception and action in ’visual form agno-sia’. Brain, 114(Pt 1B), 405–428.

Murray, I. J., & Plainis, S. (2003). Contrast coding and magno/parvo segrega-tion revealed in reaction time studies. Vision Research, 43(25), 2707–2719.

Perenin, M. T., & Vighetto, A. (1988). Optic ataxia: a speciWc disruption invisuomotor mechanisms. I. DiVerent aspects of the deWcit in reachingfor objects. Brain, 111(Pt. 3), 643–674.

Pins, D., & Bonnet, C. (1996). On the relation between stimulus intensityand processing time: Pieron’s law and choice reaction time. Perceptions& Psychophysics, 58(3), 390–400.

Robinson, D. L., & McClurkin, J. W. (1989). The visual superior colliculusand pulvinar. Review of Oculomotor Research, 3, 337–360.

Rovamo, J., Virsu, V., & Nasanen, R. (1978). Cortical magniWcation factorpredicts the photopic contrast sensitivity of peripheral vision. Nature,271(5640), 54–56.

Schiller, P. H., & Malpeli, J. G. (1978). Functional speciWcity of lateralgeniculate nucleus laminae of the rhesus monkey. Journal of Neuro-physiology, 41(3), 788–797.

Schmidt, T. (2002). The Wnger in Xight: real-time motor control by visuallymasked color stimuli. Psychological Sciences, 13(2), 112–118.

Schwartz, S. H. (1992). Reaction time distributions and their relationshipto the transient/sustained nature of the neural discharge. VisionResearch, 32(11), 2087–2092.

Toth, L. J., & Assad, J. A. (2002). Dynamic coding of behaviourally rele-vant stimuli in parietal cortex. Nature, 415(6868), 165–168.

Trappenberg, T. P., Dorris, M. C., Munoz, D. P., & Klein, R. M. (2001). Amodel of saccade initiation based on the competitive integration ofexogenous and endogenous signals in the superior colliculus. Journal ofCognitive Neuroscience, 13(2), 256–271.

Walker, R., Deubel, H., Schneider, W. X., & Findlay, J. M. (1997). EVect ofremote distractors on saccade programming: evidence for an extendedWxation zone. Journal of Neurophysiology, 78(2), 1108–1119.

Weber, H., & Fischer, B. (1994). DiVerential eVects of non-target stimuli onthe occurrence of express saccades in man. Vision Research, 34(14),1883–1891.

Wenban-Smith, M. G., & Findlay, J. M. (1991). Express saccades: is there aseparate population in humans? Experimental Brain Research, 87(1),218–222.

White, B. J., Gegenfurtner, K. R., & Kerzel, D. (2005). EVects of structurednontarget stimuli on saccadic latency. Journal of Neurophysiology,93(6), 3214–3223.

Williams, D. R., Brainard, D. H., McMahon, M. J., & Navarro, R. (1994).Double-pass and interferometric measures of the optical quality of theeye. Journal of the Optical Society of America A Optics, Image Science,and Vision, 11(12), 3123–3135.

Wright, M. J., & Johnston, A. (1983). Spatiotemporal contrast sensitivityand visual Weld locus. Vision Research, 23(10), 983–989.